EyeHint Roadmap
What EyeHint does today, what we are building next, why, and where it is going.
Dates are intentionally absent. Items move between columns as we learn. Everything under Next and Later is a plan, not a promise.
Why EyeHint exists
Most of the time you need information is exactly when you cannot stop and look it up: in the middle of a call, a demo, a lecture or a support chat. EyeHint’s job is to close the gap between what is being said or shown and what you need to know, in one glance, without interrupting the conversation and without anyone else having to see your screen.
Three principles guide every decision:
- Glanceable. If it takes more than a second to read, it is too slow.
- Private by default. On-device where possible, keys stay on your Mac, nothing stored unless you choose.
- You stay in control. EyeHint suggests; you decide what to say or do.
Use cases
| Situation | What EyeHint does |
|---|---|
| Sales and customer calls | Surfaces pricing, objections and competitor facts as they come up; drafts the follow-up email |
| Meetings | Live transcript, “what did I miss?”, action items and a summary at the end |
| Technical support and debugging | Reads an error on your screen, explains it, proposes the fix |
| Presentations and speeches | Teleprompter that follows your voice; answers audience questions fast |
| Learning and lectures | Explains a term or formula the moment it is mentioned; builds notes |
| Language and accessibility | Live captions and translation; plain-language explanations; reading support |
| Pitch and presentation practice | Mock Q&A, answer coaching, feedback on pacing and filler words |
| Research and reading | Select any area of the screen, ask a question about it |
Use EyeHint openly and where it is allowed. Tell people when you record or transcribe, and do not use it in exams or assessments that prohibit outside assistance. See the Acceptable Use Policy.
Where we are
Shipped (macOS app)
- Floating glass overlay, always on top, resizable, adjustable transparency
- Live listening to system audio (“Them”) and microphone (“You”); on-device or cloud speech-to-text
- One-tap answers, auto-answer after a pause, quick actions (summary, action items, catch-up)
- Smart follow-up chips with Explain / Why / Example / Facts and custom angles
- Screenshots, area capture, scrolling capture, trigger words, screenshot prompts
- Optional privacy mode: keep the overlay out of screen sharing and recordings, with an on/off setting
- Rebindable global shortcuts, searchable local history, presets, level-meter styles
Next — make it sellable and reliable
| Item | Why it matters |
|---|---|
| Tests, signed and notarized releases, auto-update | Trust: it must install cleanly and update safely |
| Licensing (activation, offline grace) | Needed to sell |
| EyeHint API: LLM proxy with rate limits and metering | Users get zero-setup AI; we control cost and abuse |
| Long-session controls | Meetings run 1–5 hours: rolling context, idle stop, per-session budgets |
| Playbooks: prepared answers and flows | Answers you prepare ahead are matched on-device and shown instantly with no AI request; AI is only used for the unexpected. Cheaper, faster and more accurate than asking the model every time |
| Smart teleprompter | Turns an answer into something you can read aloud naturally, and follows your voice |
| Accent themes inside the app | Matches the website and macOS accent |
Later — make it more useful
- Meeting notes: summary, decisions and action items after every call, exportable to Markdown, Notion, email
- Practice mode: a mock Q&A coach for pitches and presentations, with feedback on structure, pace and filler words
- Playbook drafting: let EyeHint draft likely questions and answers from your notes and documents for you to review, edit and approve before use
- Custom knowledge: point EyeHint at your own docs, CRM notes or product sheets so answers use your facts
- Translation and captions: live translation in the overlay
- Team features: shared prompt libraries and presets
- More platforms where the capture and privacy-mode APIs allow it
The long-term vision: an EyeHint that watches and answers
Today you ask, and EyeHint answers. The future we are working toward is an EyeHint that notices on its own what is on your screen and in your call, and has the answer ready before you ask: a question appearing in a chat, an error dialog, a form field, a slide with a number you need to explain.
This is a future goal. We will approach it in stages so each step is useful and safe on its own:
| Stage | Behavior | Control |
|---|---|---|
| 1 · Assisted (today) | You ask or tap a chip | You trigger everything |
| 2 · Proactive suggestions | EyeHint reads the screen and offers a ready answer chip when something looks answerable | You tap to see it |
| 3 · Autonomous answers | EyeHint writes the answer for what is on screen and shows it in the overlay without being asked | Per-app allowlist, pause shortcut, daily budget |
| 4 · Agent actions | EyeHint can draft replies or fill fields for you to review | Always requires your explicit approval |
Ground rules for autonomy
- Opt-in, per app. Nothing watches your screen unless you enable it for that app.
- Visible. A clear indicator whenever EyeHint is reading the screen, and a one-key pause.
- Suggest, never act silently. EyeHint will not send, click or submit anything on its own.
- Local first. Screen understanding on-device where possible; only what is needed leaves your Mac.
- Budgeted. Autonomous mode runs against a visible usage budget so it can never run away.
What has to exist first: the metered API and rate limits, reliable long-session handling, fast on-device screen understanding, and a settings model for per-app permissions. That is why they come before this on the roadmap.
Tell us what to build
Have a use case that is not listed? Write to hello@eyehint.com. The roadmap follows what people actually use.